Pareto chart

A Pareto chart is a bar chart of a problem split into categories, sorted largest first, with a line showing the running total as a percentage. It's named after Vilfredo Pareto, an economist who noticed that a small share of people held most of the land in Italy. Joseph Juran, a quality pioneer, turned that into a working rule of thumb for quality problems: a vital few causes usually account for most of the trouble, and the rest are a useful many. People call it the 80/20 rule, but that's a shorthand, not a law. Real data might split 70/30 or 90/10, or not concentrate at all. The chart's job is to show you which it is before you spend effort, so you work on the biggest part of the problem rather than the loudest or the easiest. It tells you where to look, not why it happens: the cause analysis comes next. Sources: Joseph M. Juran, Quality Control Handbook (McGraw-Hill, 1951); Kaoru Ishikawa, Guide to Quality Control (Asian Productivity Organization, 1968).

What you're counting

Write down exactly what one event is, in words someone else could count the same way: "a painted part rejected at final inspection for a finish defect", not "quality issues". This is the effect, the problem you're splitting up. If two people would count differently from your description, tighten it before collecting anything.

How you split it

Choose the axis on purpose: the one way you'll split the effect into categories. Stratifying means sorting the events into groups by one characteristic, such as defect type, cause, location, machine, shift, product or customer. Pick the split most likely to point at something you can act on, and say why. Use categories that don't overlap, so each event lands in exactly one. If you're not sure which split matters, chart two or three and keep the one that shows a clear top bar. This first split is level 1.

What you rank by

Decide what the bars measure: how often it happens (count), what it costs, or how much time it loses. Ranking by frequency and by cost can put the categories in a different order: a rare defect that scraps a whole batch can matter more than a common one fixed in seconds. Choose the measure that matches why the problem matters, write down the unit (rejected parts, dollars, minutes), and don't mix measures on one chart.

Period and data source

Use one fixed period, with a start and end date, and the same data source for every category. Mixing a month of one category with a week of another, or the day shift's log with both shifts' log, makes the ranking meaningless. Pick a period long enough to cover normal variation (often two to six weeks) and note anything unusual that happened in it. Write where the numbers came from, and how you checked them.

Categories and values

Enter each category and its total for the period. Count from the record, not from memory. Keep an Other bucket for small, one-off categories, but keep it small: if Other is among the largest bars, open it up and give its biggest items their own categories. The chart sorts the bars for you, with Other always last.

Chart and vital few (computed)

Nothing to enter here: the chart ranks the categories, works out each one's share and the cumulative percentage, and marks the vital few, the named categories that together reach roughly 80% of the total. Read it as a rough guide. If the bars are all about the same height, or it takes most of the categories to reach 80%, there's no clear vital few: try a different split or measure rather than working on everything. The warnings also flag too few categories, too few counted events, and an Other bucket that's too large.

Second-level breakdown of the top bar

Before acting, split the top bar again by a second axis. This is level 2. The level 1 chart splits the whole problem (every rejected part, by defect type); level 2 takes only the largest level 1 bar and splits that (only the colour contamination rejects, by where the old colour came from). A label like "colour contamination" isn't yet something you can fix, but its biggest source often is. Break the largest bar down first rather than moving on to the second bar, and check the level 2 values add up to the level 1 bar.

What happens next

Write what happens after the chart. A Pareto chart shows where the problem is concentrated, not why. The usual next step is a cause analysis on the top level 2 category: a fishbone diagram to lay out possible causes, and 5 Whys to follow the strongest one to something you can act on, then an A3 if it's a bounded problem worth a full record. Also note when you'll redraw the chart, with the same split, measure and period length, to see whether the countermeasure moved the top bar.

The example below continues Line 2's paint booth. With changeovers now about 35 minutes, final inspection counted 100 parts rejected for a finish defect over four weeks, from the reject tags checked against the rework log. The split is by defect type, ranked by count, with a sentence on why count is the right measure for this line (every reject costs about the same to repaint). Colour contamination is the clear top bar at 52%, and three named categories reach roughly 80%. The Other bucket is small and last. At level 2, the colour contamination rejects are split by where the old colour came from: residue in the spray gun or nozzle is 31 of the 52. The next step is a cause analysis on that residue with the booth operators, and a recount over the same length of period afterwards.

• Treating 80/20 as exact.

  The 80% line is a rule of thumb. Forcing the data to show 20% of categories causing 80% of the problem, or ignoring a chart that shows 60/40, turns a way of looking into a belief. Read what the chart actually shows, and if there's no clear vital few, say so and try another split.

• A huge Other bar.

  When Other is among the largest bars, the chart is hiding categories that may belong in the vital few. It usually means the categories were set too narrowly or the record is vague. Open the Other bucket, name its biggest items, and redraw.

• Mixing periods or sources.

  Four weeks of one defect next to one week of another, or one shift's log next to both shifts' log, ranks the data collection, not the problem. Every bar needs the same period and the same source.

• Counting the wrong thing.

  Ranking by how often something happens when cost or customer impact is what matters (or the reverse) points the team at the wrong bar. Frequency and cost can rank the same categories differently, so choose the measure on purpose and write down why.

• Charting opinions instead of counts.

  A chart of what people say the main problem is measures how memorable or annoying each issue is, not how often it happens. Count events from a record, or go and tally them first-hand.

• Stopping at the chart.

  A Pareto chart shows where to look, not why it happens. Jumping from "colour contamination is the top bar" straight to a fix skips the level 2 breakdown and the cause analysis, and the fix often lands on the wrong cause.